Title :
An Analysis of Stability of ECOC Kernel Machines
Author :
Lu, Weijun ; Chen, Hong ; Cheng, Dexin
Author_Institution :
Sch. of Comput. Sci., Wuhan Univ. of Sci. & Eng., Wuhan, China
Abstract :
Multi-class classifier is usually constructed by means of combining the outputs of several binary ones, according to an error correcting output code (ECOC) scheme. In the paper, within the framework of ECOC, we analyse of the ECOC of kernel machines originally proposed before. Then we present the generalization bounds of the ECOC of kernel machines according to the results of stability and generalization introduced in paper before.
Keywords :
error correction codes; generalisation (artificial intelligence); learning (artificial intelligence); pattern classification; ECOC kernel machines; binary ones; error correcting output code scheme; generalization bounds; multiclass classifier; Algorithm design and analysis; Decoding; Kernel; Machine learning; Stability analysis; Support vector machines; Training; Error correcting output codes(ECOC); Generalization error; Uniform stability;
Conference_Titel :
Intelligence Information Processing and Trusted Computing (IPTC), 2010 International Symposium on
Conference_Location :
Huanggang
Print_ISBN :
978-1-4244-8148-4
Electronic_ISBN :
978-0-7695-4196-9
DOI :
10.1109/IPTC.2010.34
